Automated Fleet Data Ingestion and DMS Import

Background

Carrivo is a car subscription brand in Germany, backed by a major vehicle importer and automotive group. Its fleet data arrives from OEM partners as files rather than through an API, so every new file has to be reconciled against what the business already holds before it can reach the dealer management system.

Stellantis, Carrivo's primary OEM, maintains a master Excel file of more than 600 vehicle records. Each time vehicles were purchased or existing records updated, Stellantis sent the complete file again, leaving Carrivo's operations team to work out what had changed by hand.

The challenge

• Every update arrived as a full file of 600 plus rows with no indication of what had changed
• Operations compared each new file row by row against the previous version to find new VINs and updated records
• The Movec DMS import file was then built manually to an exact specification
• VIN validation at that record volume was highly prone to human error
• Data arrived in several formats with no standard way to extract, validate or track it
• No processing history existed, so runs could not be audited or repeated with confidence

Goals

• Remove manual row by row comparison from the vehicle data process
• Detect new and updated VIN records automatically on every upload
• Validate records before export rather than after a failed import
• Produce a clean, import-ready file matched to the Movec DMS specification
• Give the team a tracked, auditable record of every processing run
• Establish a single ingestion foundation able to support further data sources

What JRNY Services delivered

Centralised ingestion module

Built the Movec DMS ingest module inside the Carrivo Fleet Ingest Platform, with one drag and drop interface accepting Excel, PDF and ZIP uploads.

Intelligent delta detection

Each upload is compared against previous uploads held in the platform, isolating only new VINs and updated records so no manual comparison is required.

Automated validation

VIN format and length checks and field completeness rules run before any output is generated.

Run level tracking

Every run is logged with total rows processed, valid rows, new VINs, updated VINs and records requiring attention, with reasons the team can drill into.

Attention export

Flagged records download as a separate file with per record error codes, so exceptions can be resolved without holding up the main import.

Import-ready output

Clean CSV formatted precisely to Movec DMS specifications, ready to import without further manual manipulation.

Technology stack

• Frontend React and TypeScript for a fast, type-safe dashboard interface
• Backend n8n Cloud Pro for workflow orchestration and API integration
• Database BigQuery on Google Cloud for record storage and delta comparison
• Infrastructure Google Cloud Frankfurt for EU data residency compliance
• Access Password protected access for simple, secure team use

The hardest part

Inconsistent input data. The Stellantis Excel format was not always consistent between updates, and records contained edge cases including partial VINs, near duplicate entries with slight variations and unexpected column ordering. Building a comparison engine robust enough to handle all of it took multiple fallback mechanisms and extensive testing against real production data.

Results

• Around 90 percent of the manual time removed from the process
• Runs that previously took significant manual effort now complete in seconds
• Data accuracy improved through automated VIN validation
• Every run fully tracked with downloadable audit records
• Three file formats supported through a single interface
• High volumes of vehicle data handled without adding headcount

Key milestones

• Unified upload interface replacing ad hoc file handling
• Delta detection engine hardened against real production edge cases
• Validation and attention reporting introduced ahead of export
• Movec DMS export format matched exactly for clean first time imports
• Run history and audit trail established across the platform

What comes next

BigQuery served the project well, though a transactional database such as PostgreSQL is the better long term choice for an application driven system of this kind, as streaming buffer limitations added complexity to the comparison logic. Future iterations will introduce more advanced anomaly detection alongside real time monitoring and alerts.